Homologous ligands accommodated by discreteconformations of a buried cavityMatthew Merski1,2, Marcus Fischer1, Trent E. Balius1, Oliv Eidam3, and Brian K. Shoichet4

Department of Pharmaceutical Chemistry, University of California, San Francisco, CA 94158-2550

Edited by Brian W. Matthews, University of Oregon, Eugene, OR, and approved March 12, 2015 (received for review January 13, 2015)

Conformational change in protein–ligand complexes is widelymodeled, but the protein accommodation expected on binding acongeneric series of ligands has received less attention. Given theiruse in medicinal chemistry, there are surprisingly few substantialseries of congeneric ligand complexes in the Protein Data Bank(PDB). Here we determine the structures of eight alkyl benzenes,in single-methylene increases from benzene to n-hexylbenzene,bound to an enclosed cavity in T4 lysozyme. The volume of theapo cavity suffices to accommodate benzene but, even with tolu-ene, larger cavity conformations become observable in the elec-tron density, and over the series two other major conformationsare observed. These involve discrete changes in main-chain con-formation, expanding the site; few continuous changes in the siteare observed. In most structures, two discrete protein conforma-tions are observed simultaneously, and energetic considerationssuggest that these conformations are low in energy relative tothe ground state. An analysis of 121 lysozyme cavity structuresin the PDB finds that these three conformations dominate the pre-viously determined structures, largely modeled in a single confor-mation. An investigation of the few congeneric series in the PDBsuggests that discrete changes are common adaptations to a seriesof growing ligands. The discrete, but relatively few, conformationalstates observed here, and their energetic accessibility, may haveimplications for anticipating protein conformational change inligand design.

conformational change | protein–ligand complexes | congeneric series |homologous series | T4 lysozyme

The importance of conformational flexibility in protein–ligandinteractions is widely acknowledged. Structural studies of

model systems such as dihydrofolate reductase (1, 2), cyclophilinA (3), adenylate kinase (4), and others (5, 6) have suggested thatconformational changes in the protein are coupled to progressalong the catalytic reaction coordinate, and that local fluctua-tions can affect coupling between binding and global transitions(7). For signal transduction, the importance of such conformationalchanges has long been recognized (8), and has been emphasized byrecent experimental (9) and computational studies (10). Accord-ingly, molecular dynamics simulations of protein–ligand complexesare now widely considered for ligand design (11–16).Despite the attention lavished on protein conformational

change overall, the incremental protein accommodations thatmight be expected over a series of ligand perturbations havereceived less consideration. Indeed, in the teeth of the “methyl,ethyl, propyl, butyl. . .futile” aphorism and the many medicinalchemistry programs that explore such incremental perturbations,surprisingly few crystal structures of congeneric ligands bound toa single protein are publicly available. Of the few there are, noneresolve decisively how a protein might accommodate a conge-neric series of ligands. If we define a congeneric series as onewith at least six ligands related through an incremental change infunctionality, then only 13 of these are known in the ProteinData Bank (PDB), and all but 2 (bold and underlined in SIAppendix, Table S4) of these undergo little conformationalchange upon ligand binding—a point to which we return. Con-versely, ligand binding leads to substantial conformationalchanges in therapeutic targets such as aldose reductase (17),

dihydrofolate reductase (2), and tRNA-guanine transglycosylase(18), but here the perturbations among the ligands have oftennot been systematic enough to disentangle changes in ligand sizeand polarity, making it harder to isolate the receptor confor-mational changes involved and their origins. In addition, similarligands can adopt dissimilar binding modes in the same protein (19).Ideally, one would like series of ligands where size and phys-

ical properties are increased incrementally without introducingother perturbations that could change binding determinants.Correspondingly, one would like a site where the growing ligandforces receptor accommodations. One system that recommendsitself is the cavity site in T4 lysozyme created by the substitutionLeu99→Ala (L99A cavity). Formed in the hydrophobic core ofthe protein, the resulting 150-Å3 cavity is sequestered from sol-vent and is almost entirely apolar. Seminal studies by Matthewsand colleagues demonstrated that this cavity can bind aryl hy-drocarbons (20, 21), and since then the cavity and related mu-tants have become model systems for ligand recognition (22–29).Contributing to this status has been the commercial availabilityof thousands of likely ligands, many closely related to one an-other. This is something that is untrue of larger, more com-plicated binding sites, where fewer likely ligands are readilyavailable, and fewer still in congeneric series.

Significance

Many medicinal chemistry programs change ligands incre-mentally to explore protein binding and to optimize bindingaffinity. How a protein accommodates such a growing ligandseries has received remarkably little structural attention. Herewe investigate eight congeneric ligands that grow by single-methylene additions, determining their protein-bound struc-tures by X-ray crystallography, to investigate how a proteinaccommodates these changes. Rather than changing confor-mation smoothly to complement the ever-larger ligands, theprotein site adopts a few discrete conformations as it expands.Inspection of the few other homologous series in the ProteinData Bank suggests that such discrete conformational adapta-tions to ligand binding are common, and may be an importantconsideration in ligand design.

Author contributions: M.M., M.F., T.E.B., and B.K.S. designed research; M.M., M.F., T.E.B.,and O.E. performed research; T.E.B. contributed new reagents/analytic tools; M.M., M.F.,T.E.B., O.E., and B.K.S. analyzed data; and M.M., M.F., T.E.B., O.E., and B.K.S. wrotethe paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.

Data deposition: The datasets reported in this paper have been deposited in the ProteinData Bank, www.pdb.org (PDB ID codes 4W51–4W59).1M.M., M.F., and T.E.B. contributed equally to this work.2Present address: Instituto de Biologia Molecular e Celular, Universidade do Porto, 4150-180 Porto, Portugal.

3Present address: pRED Informatics, Roche Pharma Research and Early Development,Roche Innovation Center Basel, 4070 Basel, Switzerland.

4To whom correspondence should be addressed. Email: bshoichet@gmail.com.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1500806112/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1500806112 PNAS | April 21, 2015 | vol. 112 | no. 16 | 5039–5044

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Dec

embe

r 14

, 202

0

Here we determine the structures of eight alkyl benzenes incomplex with the L99A cavity, including complexes with ben-zene, toluene, ethylbenzene, n-propylbenzene, sec-butylbenzene,n-butylbenzene, n-pentylbenzene, and n-hexylbenzene, as well asthe apo cavity, at resolutions ranging from 1.39 to 1.80 Å. Be-cause only benzene may be readily accommodated by the apocavity but most other members of the series bound with greateraffinity, this series seemed well-suited to exploring ligand-pro-voked conformational changes. We asked whether the cavitycontinuously adapted its conformation to these incremental en-largements in ligand size or whether instead the cavity jumped todiscrete conformational states. Using multiconformational crys-tallographic refinement, we investigated whether the new con-formations of the cavity were accessible to the ground state orwere more distant in energy. Expecting continuous changesamong conformations, we were surprised by the structures thatemerged. Comparison with other series in the PDB suggests thatthese types of protein changes may be common, with implica-tions for anticipating protein accommodation in ligand design.

ResultsStructures for Congeneric Ligands Bound to L99A. An attraction ofthe L99A cavity was that its structure had already been de-termined in complex with a partly congeneric series of ligands,including benzene, ethylbenzene, and n-butylbenzene (21). Tofill in the series, we determined structures with the missing tol-uene and n-propylbenzene and extended it with structures of sec-butylbenzene, n-pentylbenzene, and n-hexylbenzene. Structureswere determined to 1.56, 1.64, 1.63, 1.80, and 1.39 Å, respectively(SI Appendix, Table S1). Unlike the earlier members of this series,which had been collected on a home X-ray source, these datasetswere collected at a third-generation synchrotron and were refined asa multiconformer model using the program PHENIX (30, 31). Foreach complex, there was clear evidence for the ligands and sur-rounding cavity residues in unrefined Fo − Fc electron density maps(Fig. 1 and SI Appendix, Fig. S1). On refinement of the n-propyl-,sec-butyl-, n-pentyl-, and n-hexylbenzene complexes, two majorconformations of the protein were distinguishable in the regionof the “flexible” F helix of the enzyme (residues 107–115),which gates the cavity site (Fig. 1 and SI Appendix, Fig. S1).The movement of the F helix, which transitions from an

α-helix toward a 310 helix in these structures, contrasted withwhat had been observed in the earlier benzene, ethylbenzene,and n-butylbenzene structures, determined over 20 y ago. In theearlier structures, the F helix was not refined to adopt alternatestates, which is reflected by a sharp rise in crystallographic B factorsin those structures (SI Appendix, Fig. S2). This discrepancy moti-vated us to redetermine these structures afresh. We did so for theL99A protein in its apo state and in complex with benzene, ethyl-benzene, and n-butylbenzene to resolutions of 1.45, 1.50, 1.79, and1.68 Å, respectively, all at a third-generation synchrotron (SI Ap-pendix, Table S1). In the four redetermined structures, the positionsof the ligands were observed clearly in unrefined Fo − Fc electrondensity, as were the positions of surrounding cavity residues (SIAppendix, Fig. S1). Here again, two major conformations of theprotein were refined with convincing electron density (Fig. 1).

Protein Conformational Changes on Ligand Perturbation. Analysisof the nine structures, from the apo through to the complexwith n-hexylbenzene, revealed three major conformations of thecavity, two of which typically existed in the same structure (Figs.1 and 2). In the apo and in the benzene-bound structures a“closed” conformation of the cavity dominated, in which the li-gand was fully enclosed by the cavity without obvious access tobulk solvent (Fig. 2). This closed cavity occupies ∼90% of theobserved electron density; the remaining 10% was unmodeled(Figs. 1 and 2A). However, even in the toluene structure, asecond, “intermediate” conformation of the cavity, whichpartially opens to bulk solvent (Fig. 2B), was observed at 20%occupancy (Fig. 2A). In this intermediate conformation, thehydrogen-bonding pattern changes from an α-helix to a shorter

length typical of a 310 helix between the carbonyl oxygen ofThr109 and residue Ala112. Also, the hydrogen bonds from thecarbonyl oxygens of Gly107, Gly110, and Gly113 to the amidenitrogens of residues Val111, Gly113, and Thr115, respectively,present in the apo structure, are lost (SI Appendix, Fig. S3A). Thereduction in the number of backbone hydrogen bonds in theintermediate conformation suggests a higher energy state thanthe apo, closed conformation (SI Appendix, Fig. S3). This in-termediate, expanded conformation became more dominant fromthe toluene to n-butylbenzene cavity complexes, ranging from 20 to60% occupancy, although it continued to coexist with the closedconformation of the cavity in its ligand-bound state (Fig. 2). In then-butylbenzene complex a third “open” conformation of the cavityappeared (30% occupancy), coexisting with the intermediate andclosed conformations at 60 and 10% occupancy, respectively (Fig.2). In this open conformation, the 310-like hydrogen bond betweenGlu108 and Val111 is maintained from the intermediate conforma-tion, whereas the hydrogen bonds present in the closed conformationfrom the amides of Gly113 and Thr115 reappear, although decreasedand increased in connectivity by one residue, respectively (SI Ap-pendix, Fig. S3). The open conformation, which dominates then-pentylbenzene and n-hexylbenzene structures, has two fewer intra-protein hydrogen bonds than the closed conformation, and is nowopen to bulk solvent (Fig. 2B), exposing substantially more hydro-phobic surface area than either the closed or the intermediate states,consistent with a higher energy conformation (SI Appendix, Fig. S3).

Discrete Protein Conformational States.Quantitative analysis of thenine structures supports their clustering into three discrete pro-tein conformations. We measured the rms deviations among themain-chain atoms of the nine structures for F-helix residues 107–115, the only region of the protein that underwent substantial

Fig. 1. Electron density maps reveal alternative F-helix conformations in theL99A cavity (stereoview). For illustration, only the major conformation of theF helix and of the ligands was refined (2mFo − DFc maps as blue mesh, 1σ).The stick size corresponds to relative occupancies to which the three alter-native F-helix conformations ultimately refined: closed (purple), inter-mediate (cyan), and open (green), which coexist in the individual complexes.When the major F-helix conformation was refined at 100% occupancy, dif-ference electron density (mFo − DFc maps, ±3σ) appears for missing alter-native conformations (green mesh), whereas partial occupancy of the majorF-helix conformation is indicated by negative density (red mesh). (A) In theethylbenzene complex, the presence of the intermediate conformation ofthe F helix, and a second conformation of the ligand, is indicated by positiveelectron density (green mesh). Correspondingly, the closed conformation ofthe F helix, refined at 100% occupancy, is associated with negative electrondensity (red mesh); both conformations ultimately refined to around 50%(cf. Fig. 2). (B) In the n-hexylbenzene complex, the difference density mapssupport the presence of the closed conformation, which cannot accommo-date the large ligand and is modeled as apo and the open conformation ofthe F helix; these ultimately refined to 30% and 70%, respectively.

5040 | www.pnas.org/cgi/doi/10.1073/pnas.1500806112 Merski et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 14

, 202

0

movement (Fig. 3 A and B). One conformation observed in eachof the nine structures adopted a closed conformation, with rmsdvalues among the main-chain atoms within 1.4 Å of one another.For the apo structure and the benzene and toluene complexesthis conformation dominated, whereas in the ethyl-, n-propyl-,and sec-butylbenzene structures, this closed conformation isabout equiprevalent with the intermediate conformation (Fig.2A). In the structures with the largest ligands, the closed con-formation, which can still be observed as a minor component,likely represents the percentage of protein molecules that areunoccupied by ligand, and so are in fact in an apo state (SIAppendix, Table S6). Correspondingly, the intermediate confor-mation of the protein could also be clustered by main-chain rmsdvalues of typically 1.0 Å or less among all five structures in whichit appeared. The three open conformations of the F helix alsodiffered by no more than 1.0 Å (Fig. 3B). Conversely, betweenthe closed, intermediate, and open states, structures differedfrom one another by rmsd values of at least 1.8 Å, typically more(Fig. 3 A and B). Both visually (Fig. 3A) and by measurement(Fig. 3B), the closed, intermediate, and open states may bedistinguished, and represent jumps without substantial in-termediates between them (see below for a partial exception).This clustering of states based on main-chain atoms could bereplicated at the side-chain level. For instance, Val111, whichin previous structures has been modeled to occupy multiplerotamers (24), also occupied three distinct states in the ninestructures (Fig. 3 C and D). Whereas within each state slightvariations in the position of Val111 were observed, these wereuncorrelated with ligand size and seemed to differ little andwithout trend; the differences in the position of Val111 amongthe three states were unambiguous.An exception to the occupancy of discrete states was the be-

havior of the main-chain atoms of Glu108 at the hinge region ofthe F helix. Whereas in the closed conformations these atoms

cluster to occupy a single conformation (Fig. 3 E and F), acontinuous and relatively smooth change is observed in theintermediate and open complexes, as the cavity expands to ac-commodate larger and larger ligands (Fig. 3 E and F). This hingeregion is the one area of the structure where conformationsvaried smoothly as the ligands grew in size.

Energy of Ligand Binding and Conformational Strain. The appear-ance of multiple conformational states in a single structuresuggests that the alternate conformations are relatively low-energy and accessible. To set boundaries on how high in energythese states might be, the ligand binding energies were comparedwith what might be expected if ligands were binding optimally,and not paying a cost for protein conformational change. Asshown by Morton and Matthews in their seminal calorimetricstudy (21), as ligands grow from benzene to toluene, ethyl-,n-propyl-, and n-butylbenzene, affinity rises linearly, but at onlyhalf the pace of the water→octanol transfer free energies (Table1) (the affinities of n-pentyl- and n-hexylbenzene were in-accessible owing to solubility limits). For instance, toluene bindsto L99A with a Kd of 100 μM (ΔGbind of −5.2 kcal/ mol),0.3 kcal/mol better (lower) than does benzene (Kd of 175 μM,ΔGbind of −5.5 kcal/mol) (21). Meanwhile, by water→octanoltransfer energies, one might expect toluene to bind 0.8 kcal/molbetter (20, 32), a difference between observed and expected af-finities of 0.5 kcal/mol (Table 1). Similarly, n-butylbenzene binds1.5 kcal/mol better than does benzene, whereas water→octanoltransfer free energies would have it binding 2.9 kcal/mol better(20, 32). Several reasons for this difference in observed andexpected binding energies have been mooted, including a proteinreorganization energy (21). If we attributed all of the discrepancybetween the observed ΔΔGbind and that predicted by transferfree energies to the cost of changing conformations from the closedto the intermediate to the open states, then the maximum energyfor the intermediate conformation, represented by the ethyl- and n-propylbenzene complexes, and for the open conformation, repre-sented by the n-butylbenzene complex, cannot exceed 0.8 and1.4 kcal/mol, respectively (Table 1). Thus, both the co-occupancy ofthe conformations in individual crystal structures and the retentionof substantial ligand affinity for the larger protein cavity confor-mations suggest that the intermediate and open conformationsrepresent relatively low-energy, accessible states.

Conformations in Other Lysozyme Cavities Recapitulate the ThreeStates. To investigate the generality of the conformations ob-served among the nine structures determined here, we analyzed121 lysozyme cavity structures in the PDB (including apoand ligand-complexed structures of L99G, L99A/M102Q, L99A/M102H, L99A/M102E, L99A/M102L, L99A/F153A, L99G/E108V,and L99A/E108V). With the exception of 11 of these, all of thesestructures were refined to represent a single structure, notallowing for multiple conformations to be occupied. Intriguingly,based on the rms deviations of the F helix, the same three statesdominated these deposited structures as well (Fig. 3G and SIAppendix, Figs. S4–S7 and Tables S2 and S3). Eighty of thesestructures occupied the closed conformation, another 28 occu-pied the intermediate conformation, and 4 occupied the openconformation (SI Appendix, Figs. S4 and S5 and Table S2).Finding our three conformations in these independently de-termined structures attests to their accessibility and occupancy inresponse to ligands beyond our congeneric series.

Conformational Accommodations in Other Protein–Ligand Series. Toinvestigate whether the discrete protein accommodations to thecongeneric ligands are common or unusual, we searched thePDB for other ligand series (Methods). We sought series whereeach ligand was closely related to at least one other by a hightopological similarity, and where there were six or more suchligands binding to a common protein; we insisted on six congenericligands to ensure that there were enough to observe trends. Weapplied two different criteria to select these series: (i) We looked at

Fig. 2. Congeneric ligands are accommodated in L99A with conformationalchanges. (A) In the L99A cavity, the ligand poses were assigned to their re-spective protein conformations by matching the ligand occupancy with thatof the F-helix conformation, which was typically unambiguous. (B) Molecularsurface of the cavity, cut away to reveal the ligand (orange space-fillingmodel), in examples of the closed (benzene complex), intermediate (ethyl-benzene complex), and open (n-hexylbenzene complex) conformations. Thefull congeneric series is shown.

Merski et al. PNAS | April 21, 2015 | vol. 112 | no. 16 | 5041

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Dec

embe

r 14

, 202

0

the size of a linked list of ligands differing only by one heavy atom;and (ii) we weakened the linked list criteria and added a literaturecitation requirement. Because topological similarity is sensitive toligand size (33, 34), we used the Tversky coefficient for fragments

(70–250 Da) and the Tanimoto coefficient for larger molecules(70–500 Da), each with a minimum similarity coefficient of 0.6. Toidentify those series that were actually congeneric, we furtherinsisted that they differed incrementally and linearly, interrogatingeach series closely (SI Appendix, Table S4). To our surprise, therewere only 13 proteins that had congeneric series of six or more li-gands in the PDB (SI Appendix, Table S4, underlined entries).With the exceptions of the enzymes enoyl-ACP reductase

(FabI) and Arg:Gly amidinotransferase, inspection of the proteinstructures reveals little monotonic response on the part of theprotein to ligand binding. Often this reflected binding of the li-gands on the surface of their proteins, allowing the ligand togrow into unfilled areas of the site and solvent (35–39). In thecase of FabI, the enzyme responds to a series of six side-chainelongations of diphenyl ethers with a smooth shift of Ile207 and0.5–0.9 Å movements of Tyr147 and Val201 (40) (SI Appendix,Fig. S8E). This is an example, then, of smooth side-chain changein response to a ligand perturbation, in contrast to the discretechanges we observed in the lysozyme cavity. For Arg:Gly ami-dinotransferase, there is evidence for both a smooth and discretetransition. For the discrete change the protein is in one state tobind glycine, γ-aminobutyric acid, and δ-aminovaleric acid; thereis a conformational transition when it binds norvaline, alanine, orα-aminobutyric acid, with a corresponding gain of several hy-drogen bonds (SI Appendix, Fig. S10).When we relax our criteria, we find several other proteins that

respond to smaller congeneric series of ligands with conforma-tional accommodation. We analyzed four cases in detail: sialidaseNanB (five congeneric ligands with structures), the estrogen re-ceptor (two sets of four congeneric ligands), heat shock proteinHSP90 (three congeneric ligands), and dihydroorotate de-hydrogenase (three congeneric ligands) (SI Appendix, Table S5).In each, the protein responds to growing ligands with discreteconformational accommodations. In NanB (41), the Ile350–Asn353 loop opens from a closed state upon binding 2-[(3-bromobenzyl)amino]ethanesulfonic acid and 2-[(4-methoxybenzyl)amino]ethanesulfonic acid (PDB ID codes 4FPE and 4FPY).Intriguingly, difference electron density for the structures ofPDB ID codes 4FPY, 4FPE, and 4FQ4 suggests that both closedand open loop states are present in the respective other structure(SI Appendix, Fig. S8B). The presence of the alternative con-formations in any single structure is not modeled in the de-posited data; instead, as in the early L99A complexes, the Bfactors of this region have been allowed to rise to values over90 Å2 (SI Appendix, Fig. S8C). Similarly, the estrogen receptorresponds to the larger benzoxathiins and chromanes (PDB IDcodes 1XPC, 1SJ0, and 1YIM) with a more open, tamoxifen-likeconformation of helices h3, h11, and h12, forming an antagonist-like conformation. A more closed conformation is observed for asmaller ligand series that was designed to probe the impact ofdynamic ligand binding on cellular signaling pathways (42). InHSP90, accommodations are more entangled, as here all of theligands are fragments, sometimes binding in different configu-rations, and although certainly related they are not congeneric(apart from purine-based inhibitors such as PU3, PU8, andPU9). Nevertheless, it is germane to note that the divergenthelical region from Leu103 to Lys116 exhibits three discreteconformations: a closed conformation for ligands such as ade-nine, a more open state for 9-ethyl-9H-purine-6-ylamine, andagain more open for more decorated purines such as the congenericseries mentioned above. Finally, discrete states may be observed fordihydroorotate dehydrogenase around Pro131 (residues 128–139)upon binding a series of tetrahydropyrimidine carboxylic acids.Rather than responding to changes in the homologous series, itappears that a 40–50° rotation of the ligand’s head group (e.g.,dimethoxyphenyl in ligand W75 of PDB ID code 3W75) pop-ulates an alternative, open state, reflected by a change in thehydrogen-bond pattern of the protein backbone (SI Appendix,Fig. S11). In the difference electron density maps, we found evi-dence that those states not only exist within separate chains of the

Fig. 3. Ligands are accommodated by mostly discrete conformational shiftsof the L99A cavity. Structural superpositions (A, C, and E) and rmsd heatmaps (B, D, and F) are shown for a ligand-responsive loop region (107–115; Aand B) harboring Val111 (C and D) and Glu108 (E and F). Whereas most ofthe movement may be characterized as discrete (C and D), the Glu108transitions (E and F) are at least partly smooth. Matrices B, D, and F aresorted by ligand size (closed, intermediate, and open; from small to largeligand size). Red indicates an rmsd value less than 2 Å, whereas blue indicatesan rmsd value greater than 2 Å. (G) The three conformations of the cavity loopregion (residues 107–115) are recapitulated among 121 crystal structures ofcavity variants previously deposited in the PDB. The standard variations (Å) inthe closed (purple), intermediate (cyan), and open (green) conformations inthese structures are represented by tube width around the coordinate meansof the clusters obtained by single-linkage hierarchical clustering.

5042 | www.pnas.org/cgi/doi/10.1073/pnas.1500806112 Merski et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 14

, 202

0

homodimer, as modeled, but that both states also coexist withinone of the monomers (SI Appendix, Fig. S8F).

DiscussionProtein conformational change upon ligand binding has receivedmuch attention (43, 44); what sets this study apart is its focus onprotein accommodations to a congeneric series of ligands. Asteady perturbation of molecular properties is widely practiced inligand design but has surprisingly few examples in publishedstructures. Three principal observations emerge. First, to a con-generic series of eight alkyl benzenes, the lysozyme L99A cavityresponds by discrete conformational changes, transiting amongthree states. Within each state there were small variations, revealingno steady pattern, whereas the closed, intermediate, and openconformations of the cavity were readily distinguished and repre-sented clear responses to ligand enlargement. Second, in multi-conformational crystallographic refinement, two of these threestates coexisted in most of the single complexes determined. This,and consideration of the ligand affinities, suggests that these alter-native conformations represent not only discrete but also low-energy conformations. Third, a review of PDB structures suggeststhat discrete protein accommodations to congeneric ligands are notuncommon, transiting among relatively few apparently low-energyprotein conformations, as far as we can determine given the fewsystematic cases known. These observations have implications foranticipating protein accommodations in ligand design.Both structural analyses and binding energy support the idea

that the lysozyme cavity responds with discrete, low-energy con-formational changes as the alkyl benzenes grow in size. The con-formations adopted by the F helix, as it responds to the ligands, fallinto three discernible groups of main-chain conformations (Fig. 3).Within each of these groups are small variations, but these do notthemselves track with an increase in ligand size. What does trackwith increasing ligand size is the discrete and progressive opening ofthe cavity in three clusters of conformations. The energetic acces-sibility of the more open conformations is supported by two linesof evidence. First, two or more conformations coexist with oneanother in all but one of the complexes determined here. Sec-ond, if one attributes all of the “missing” ligand binding energy,expected from increasing hydrophobicity, to protein reorganization(Table 1), then the cost of accessing the open conformation is notmore than 1.4 kcal/mol; we consider caveats to this argument below.If the closed, intermediate, and open conformations are ac-

cessible conformational states of the lysozyme cavity, then theyshould be seen in other cavity structures, responding to differentperturbations. Consistent with this view, 121 previously pub-lished lysozyme cavity structures, representing cavity variants andtheir complexes with multiple ligands, could be readily clusteredinto the same three conformational substates represented by thenine structures determined here (SI Appendix, Fig. S4). Theseearlier structures typically represent single structure refinementsthat nevertheless recapitulate the same major states.Similarly, if discrete protein accommodation to congeneric li-

gands is common, then it should be observable in other protein–

ligand structures. Although we were surprised at just how fewsubstantial congeneric series are represented structurally, thefew there are often do undergo such discrete conformationalchanges. Thus, NanB, the estrogen receptor, dihydroorotatedehydrogenase, HSP90, and Arg:Gly amidinotransferase all re-spond to congeneric series of ligands with discrete conforma-tional accommodations. As with the L99A cavity, these discreteconformations could often be observed in the same crystalstructure when the electron density is inspected, even if, as wasthe case with NanB, they were originally modeled as having onlya single, typically high B-factor conformation (21, 41). Naturally,we do not pretend that such discrete conformational accom-modations are the only ways for a protein to respond to conge-neric ligands; there were also several cases where the proteinsresponded to the ligands relatively smoothly, as was the case forFabI and Src kinase, and several congeneric series could be ac-commodated without movement of the protein at all. Still, themovement observed in the protein was often a substantial, discreteone, as though jumping from one low-energy state to another.Certain caveats merit airing. Even in the L99A cavity, there

are movements that can be represented as smooth rather thandiscrete. For the motions of the main-chain atoms of Glu108, onthe periphery of the F helix, the distinction between the openand intermediate conformations collapses into a smooth ac-commodation to larger ligands (Fig. 3 E and F); the closedconformation could still be distinguished as a separate state.Neither do we pretend to have undertaken a comprehensivestudy of protein accommodations to small ligand changes, althoughwe do hope to have captured the responses to most congenericseries. In many complexes, a small ligand perturbation will lead to acorrespondingly small change in the protein until a tipping point isreached that pushes the protein into discrete conformationalchange, to occupy what appear to be a relatively small number ofalternate, low-energy states. The inference that the intermediateand open conformational states are low-energy, based on differen-tial ligand binding energies, depends on what one expects forthe binding energies unencumbered by protein conformationalstrain. If optimal binding affinity should exceed the contribution ofincreased ligand hydrophobicity, then the inferred conformationalstrain would rise. However, even if one insists on an atomic ligandefficiency of 1 kcal/mol per added methylene (the ligand efficiency ofbenzene in the L99A cavity is 0.85), then n-butylbenzene should bind4 kcal/mol better than benzene, rather than the 1.5 kcal ΔΔG ac-tually observed. If one attributed that difference entirely to proteinconformational strain, then the energy difference between the closedstate and the more open states would still only be 2.5 kcal/mol.Ligand optimization is the stage on which most effort is lav-

ished in drug design and probe development. In such optimiza-tion, ligands are often perturbed incrementally. It is natural toassume that proteins will respond with correspondingly smallaccommodations, which may be modeled by techniques withsmall radii of convergence, such as structure relaxation and shortmolecular dynamics simulations. This study suggests that rela-tively large, discrete changes in protein conformation to smallligand perturbation may be common. Anticipating such changesmay demand long molecular dynamics simulations (10, 45–48) orapproaches that sample among precalculated states (49, 50). Thelow energy of these states would also support a combination ofexperimental observation, such as NMR or crystallographic re-finement, with computational modeling of protein conforma-tional changes in ligand optimization (31, 51). Irrespective ofmethod, modeling discrete jumps among relatively few low-energy protein states may often be an important considerationin the structure-based design of new ligands.

MethodsProtein Crystallization and Structure Refinement. T4 lysozyme/L99A wascloned, purified, and crystallized as described (SI Appendix, Methods).Datasets on cryocooled crystals were collected at the Advanced Light Sourcebeamline 8.3.1 and processed with XDS (52) and Phaser (53). To removemodel bias from molecular replacement model 181L, F-helix residues 107–115

Table 1. Ligand affinity and protein conformation

Ligand Kd, μM*ΔΔGbind,

kcal/mol*,†,‡ΔΔGwat,oct,kcal/mol*,†

Occupancy, %

C I O

Benzene 175 0.0 0.0 90 — —

Toluene 102 −0.33 −0.82 80 20 —

Ethylbenzene 68 −0.57 −1.40 50 50 —

n-Propylbenzene 18 −1.36 −2.13 60 40 —

n-Butylbenzene 14 −1.50 −2.91 10 60 30

C, closed; I, intermediate; O, open loop conformation.*From refs. 20 and 35.†ΔΔG with reference to benzene.‡The ΔG of benzene is −5.19 ± 0.16 kcal/mol.

Merski et al. PNAS | April 21, 2015 | vol. 112 | no. 16 | 5043

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Dec

embe

r 14

, 202

0

and the ligand were excluded from the starting model and added in the laterrounds of model building, when occupancies were refined automatically usingPhenix.refine (30) applying a 10% cutoff for modeling alternative conforma-tions (cf. Fig. 2A). The presence of additional conformations in a structure wasevidenced by features in the Fo − Fc density maps (σ = 3.0) and decreases inRfree upon modeling additional conformations. Datasets were depositedin the PDB as 4W51 (no ligand), 4W52 (benzene), 4W53 (toluene), 4W54(ethylbenzene), 4W55 (n-propylbenzene), 4W56 (sec-butylbenzene), 4W57(n-butylbenzene), 4W58 (n-pentylbenzene), and 4W59 (n-hexylbenzene).

PDB Search for Other Homologous Series. We included protein systems thatmet the following criteria: (i) ≥6 linked ligands with Tanimoto index >0.6,and (ii) have a molecular formula difference of ≤1 heavy atom. As an al-ternative, we replaced criteria (ii) with ≥6 ligands associated with one paper.

We performed this procedure for both ligands (70–500 Da) and fragments(70–250 Da) and calculated a (weighted) Tanimoto coefficient with a cutoffof 0.6. Corresponding lists of (i) 65 and (ii) 120 structures (40 overlap) werehand-curated for references, ligand similarity, and PDB structures. More detailsare in SI Appendix, Methods; Python and csh scripts are available upon request.

For the analysis of 121 PDB-deposited lysozyme cavity structures and forour 9 structures, we used UCSF Chimera (54) for the alignment and rmsdcalculation of the F helix (residues 107–115). Hierarchical clustering analysiswas performed and heat maps were generated using Python 2.7.

ACKNOWLEDGMENTS.We thank Matthew J. O’Meara and Sarah Barelier forreading this manuscript. For enlightening conversations on the lysozymecavity, B.K.S. thanks Andy Morton (1964–1997). Supported by GM59957 (toB.K.S.) and National Research Service Award F32GM108161 (to T.E.B.).

1. Chen S, et al. (2013) Detection of dihydrofolate reductase conformational change byFRET using two fluorescent amino acids. J Am Chem Soc 135(35):12924–12927.

2. Schnell JR, Dyson HJ, Wright PE (2004) Structure, dynamics, and catalytic function ofdihydrofolate reductase. Annu Rev Biophys Biomol Struct 33:119–140.

3. Fraser JS, et al. (2009) Hidden alternative structures of proline isomerase essential forcatalysis. Nature 462(7273):669–673.

4. Henzler-Wildman KA, et al. (2007) Intrinsic motions along an enzymatic reactiontrajectory. Nature 450(7171):838–844.

5. Kokkinidis M, Glykos NM, Fadouloglou VE (2012) Protein flexibility and enzymaticcatalysis. Adv Protein Chem Struct Biol 87:181–218.

6. Henzler-Wildman K, Kern D (2007) Dynamic personalities of proteins. Nature450(7172):964–972.

7. Whitten ST, García-Moreno E B, Hilser VJ (2005) Local conformational fluctuations canmodulate the coupling between proton binding and global structural transitions inproteins. Proc Natl Acad Sci USA 102(12):4282–4287.

8. Elgeti M, et al. (2013) Precision vs flexibility in GPCR signaling. J Am Chem Soc 135(33):12305–12312.

9. Chung KY, et al. (2011) Conformational changes in the G protein Gs induced by the β2adrenergic receptor. Nature 477(7366):611–615.

10. Dror RO, et al. (2013) Structural basis for modulation of a G-protein-coupled receptorby allosteric drugs. Nature 503(7475):295–299.

11. Cole DJ, Tirado-Rives J, Jorgensen WL (2015) Molecular dynamics and Monte Carlosimulations for protein-ligand binding and inhibitor design. Biochim Biophys Acta1850(5):966–971.

12. Hsu YH, et al. (2013) Fluoroketone inhibition of Ca(2+)-independent phospholipaseA2 through binding pocket association defined by hydrogen/deuterium exchangeand molecular dynamics. J Am Chem Soc 135(4):1330–1337.

13. Ranganathan A, Dror RO, Carlsson J (2014) Insights into the role of Asp79(2.50) in β2adrenergic receptor activation from molecular dynamics simulations. Biochemistry53(46):7283–7296.

14. Young T, Abel R, Kim B, Berne BJ, Friesner RA (2007) Motifs for molecular recognitionexploiting hydrophobic enclosure in protein-ligand binding. Proc Natl Acad Sci USA104(3):808–813.

15. Singh N, Frushicheva MP, Warshel A (2012) Validating the vitality strategy for fightingdrug resistance. Proteins 80(4):1110–1122.

16. Singh N, Warshel A (2010) Absolute binding free energy calculations: On the accuracyof computational scoring of protein-ligand interactions. Proteins 78(7):1705–1723.

17. Sotriffer CA, Krämer O, Klebe G (2004) Probing flexibility and “induced-fit” phe-nomena in aldose reductase by comparative crystal structure analysis and moleculardynamics simulations. Proteins 56(1):52–66.

18. Brenk R, Stubbs MT, Heine A, Reuter K, Klebe G (2003) Flexible adaptations in thestructure of the tRNA-modifying enzyme tRNA-guanine transglycosylase and theirimplications for substrate selectivity, reaction mechanism and structure-based drugdesign. ChemBioChem 4(10):1066–1077.

19. Mattos C, Rasmussen B, Ding X, Petsko GA, Ringe D (1994) Analogous inhibitors ofelastase do not always bind analogously. Nat Struct Biol 1(1):55–58.

20. Morton A, Baase WA, Matthews BW (1995) Energetic origins of specificity of ligandbinding in an interior nonpolar cavity of T4 lysozyme. Biochemistry 34(27):8564–8575.

21. Morton A, Matthews BW (1995) Specificity of ligand binding in a buried nonpolarcavity of T4 lysozyme: Linkage of dynamics and structural plasticity. Biochemistry34(27):8576–8588.

22. Boyce SE, et al. (2009) Predicting ligand binding affinity with alchemical free energymethods in a polar model binding site. J Mol Biol 394(4):747–763.

23. Wei BQ, Baase WA, Weaver LH, Matthews BW, Shoichet BK (2002) A model bindingsite for testing scoring functions in molecular docking. J Mol Biol 322(2):339–355.

24. Mobley DL, Chodera JD, Dill KA (2007) The confine-and-release method: Obtainingcorrect binding free energies in the presence of protein conformational change.J Chem Theory Comput 3(4):1231–1235.

25. Wang K, Chodera JD, Yang Y, Shirts MR (2013) Identifying ligand binding sites andposes using GPU-accelerated Hamiltonian replica exchange molecular dynamics.J Comput Aided Mol Des 27(12):989–1007.

26. Nucci NV, Fuglestad B, Athanasoula EA, Wand AJ (2014) Role of cavities and hydrationin the pressure unfolding of T4 lysozyme. Proc Natl Acad Sci USA 111(38):13846–13851.

27. Ucisik MN, Zheng Z, Faver JC, Merz KM (2014) Bringing clarity to the prediction of protein-ligand binding free energies via “blurring.” J Chem Theory Comput 10(3):1314–1325.

28. Lakkaraju SK, Raman EP, Yu W, MacKerell AD, Jr (2014) Sampling of organic solutes inaqueous and heterogeneous environments using oscillating excess chemical poten-tials in grand canonical-like Monte Carlo-molecular dynamics simulations. J ChemTheory Comput 10(6):2281–2290.

29. Bouvignies G, et al. (2011) Solution structure of a minor and transiently formed stateof a T4 lysozyme mutant. Nature 477(7362):111–114.

30. Adams PD, et al. (2010) PHENIX: A comprehensive Python-based system for macro-molecular structure solution. Acta Crystallogr D Biol Crystallogr 66(Pt 2):213–221.

31. Fischer M, Coleman RG, Fraser JS, Shoichet BK (2014) Incorporation of protein flexi-bility and conformational energy penalties in docking screens to improve liganddiscovery. Nat Chem 6(7):575–583.

32. Sangster J (1989) Octanol-water partition coefficients of simple organic compounds.J Phys Chem Ref Data 18(3):1111–1227.

33. Flower DR (1998) On the properties of bit string-based measures of chemical simi-larity. J Chem Inf Comput Sci 38(3):379–386.

34. Wang Y, Bajorath J (2009) Development of a compound class-directed similarity co-efficient that accounts for molecular complexity effects in fingerprint searching.J Chem Inf Model 49(6):1369–1376.

35. Biela A, et al. (2012) Impact of ligand and protein desolvation on ligand binding tothe S1 pocket of thrombin. J Mol Biol 418(5):350–366.

36. Eidam O, et al. (2012) Fragment-guided design of subnanomolar β-lactamase in-hibitors active in vivo. Proc Natl Acad Sci USA 109(43):17448–17453.

37. Mecinovi�c J, et al. (2011) Fluoroalkyl and alkyl chains have similar hydrophobicities in bindingto the “hydrophobic wall” of carbonic anhydrase. J Am Chem Soc 133(35):14017–14026.

38. Rauh D, Klebe G, Stubbs MT (2004) Understanding protein-ligand interactions: Theprice of protein flexibility. J Mol Biol 335(5):1325–1341.

39. Krimmer SG, Betz M, Heine A, Klebe G (2014) Methyl, ethyl, propyl, butyl: Futile butnot for water, as the correlation of structure and thermodynamic signature shows in acongeneric series of thermolysin inhibitors. ChemMedChem 9(4):833–846.

40. Chang A, et al. (2013) Rational optimization of drug-target residence time: Insightsfrom inhibitor binding to the Staphylococcus aureus FabI enzyme-product complex.Biochemistry 52(24):4217–4228.

41. Brear P, Telford J, Taylor GL, Westwood NJ (2012) Synthesis and structural charac-terisation of selective non-carbohydrate-based inhibitors of bacterial sialidases.ChemBioChem 13(16):2374–2383.

42. Srinivasan S, et al. (2013) Ligand-binding dynamics rewire cellular signaling via es-trogen receptor-α. Nat Chem Biol 9(5):326–332.

43. Cui Q, Karplus M (2008) Allostery and cooperativity revisited. Protein Sci 17(8):1295–1307.44. Gunasekaran K, Ma B, Nussinov R (2004) Is allostery an intrinsic property of all dy-

namic proteins? Proteins 57(3):433–443.45. Dror RO, Dirks RM, Grossman JP, Xu H, Shaw DE (2012) Biomolecular simulation: A

computational microscope for molecular biology. Annu Rev Biophys 41:429–452.46. Lane TJ, Bowman GR, Beauchamp K, Voelz VA, Pande VS (2011) Markov state model

reveals folding and functional dynamics in ultra-long MD trajectories. J Am Chem Soc133(45):18413–18419.

47. Lane TJ, Shukla D, Beauchamp KA, Pande VS (2013) To milliseconds and beyond:Challenges in the simulation of protein folding. Curr Opin Struct Biol 23(1):58–65.

48. Fenley AT, Muddana HS, Gilson MK (2012) Entropy-enthalpy transduction caused byconformational shifts can obscure the forces driving protein-ligand binding. Proc NatlAcad Sci USA 109(49):20006–20011.

49. Das A, et al. (2014) Exploring the conformational transitions of biomolecular systemsusing a simple two-state anisotropic network model. PLOS Comput Biol 10(4):e1003521.

50. Lei M, Zavodszky MI, Kuhn LA, Thorpe MF (2004) Sampling protein conformationsand pathways. J Comput Chem 25(9):1133–1148.

51. Pfleger C, Gohlke H (2013) Efficient and robust analysis of biomacromolecular flexi-bility using ensembles of network topologies based on fuzzy noncovalent constraints.Structure 21(10):1725–1734.

52. Kabsch W (2010) XDS. Acta Crystallogr D Biol Crystallogr 66(Pt 2):125–132.53. McCoy AJ, et al. (2007) Phaser crystallographic software. J Appl Crystallogr 40(Pt 4):

658–674.54. Pettersen EF, et al. (2004) UCSF Chimera—A visualization system for exploratory re-

search and analysis. J Comput Chem 25(13):1605–1612.

5044 | www.pnas.org/cgi/doi/10.1073/pnas.1500806112 Merski et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 14

, 202

0